the intergenerational transmission of human capital and ...ftp.iza.org/dp10300.pdf · discussion...

Forschungsinstitut zur Zukunft der ArbeitInstitute for the Study of Labor

DI

SC

US

SI

ON

P

AP

ER

S

ER

IE

S

The Intergenerational Transmission of Human Capital and Earnings in Contemporary Russia

IZA DP No. 10300

October 2016

Gleb V. BorisovChristopher A. Pissarides

The Intergenerational Transmission of

Human Capital and Earnings in Contemporary Russia

Gleb V. Borisov St. Petersburg State University

Christopher A. Pissarides London School of Economics and IZA

Discussion Paper No. 10300 October 2016

IZA

P.O. Box 7240 53072 Bonn

Germany

Phone: +49-228-3894-0 Fax: +49-228-3894-180

E-mail: [email protected]

Any opinions expressed here are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but the institute itself takes no institutional policy positions. The IZA research network is committed to the IZA Guiding Principles of Research Integrity. The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit organization supported by Deutsche Post Foundation. The center is associated with the University of Bonn and offers a stimulating research environment through its international network, workshops and conferences, data service, project support, research visits and doctoral program. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.

IZA Discussion Paper No. 10300 October 2016

ABSTRACT

The Intergenerational Transmission of Human Capital and Earnings in Contemporary Russia*

We make use of longitudinal data for the Russian economy over 1994-2013 to obtain earnings and education information about parents and children. We estimate the intergenerational transmission of educational attainment and earning capacity and find high intergenerational correlation of earnings for both sons and daughters independently of educational qualifications. We attribute them to the impact of informal networks. We also find high correlation of educational qualifications but with critical variations due to labour market conditions. At the time of transition around 1990 children’s educational attainment fell well below parents but recovered a decade later when the economy was booming. JEL Classification: J21, J23, J24, J62, O15 Keywords: human capital, intergenerational education mobility,

intergenerational earnings elasticity, Russia Corresponding author: Christopher A. Pissarides Department of Economics London School of Economics Houghton Street WC2A 2AE London United Kingdom E-mail: [email protected]

* This paper was written at the Growth Laboratory of the St.-Petersburg State University. We are grateful to Otar Marganiya, the Dean of the Business School, for his support and to the other members of the Laboratory for helpful interactions.

1. Introduction

This paper makes use of panel data to study the intergenerational trans-

mission of human capital and earnings in post-reform Russia. Our approach

follows that of Becker and Tomes (1979), who explain the reasons for the per-

sistence of between-family earnings differentials in terms of educational in-

vestments. They argue that higher-income families endow their children with

more human capital, and so the children become higher-earning individuals

themselves. Our main findings differ with respect to the main transmission

mechanism, in that we find that the preservation of higher earnings within

families takes place through other informal channels and not through larger

educational investments by higher-income families. We find that highly edu-

cated parents endow their children with more education, and so both parents

and children earn higher incomes; but controlling for education, higher earn-

ing parents do not endow their children with more education and yet their

children do end up being high earners themselves. Our data source is the

Russian Longitudinal Monitoring Survey (RLMS), which contains a wealth

of information about individuals and their families.1

Our study takes place in the volatile background of post-reform Russia

and we are able to identify the impact of macro fluctuations on human cap-

ital and earnings. In the last three decades Russia experienced the collapse

of the planned economy of the previous sixty years, a subsequent deep struc-

tural and financial crisis that lasted until the end of the 1990s and a rapid

growth of free-market institutions and practices during the 2000s. These

transitions were accompanied by dramatic fluctuations in the demand for

skilled labour by firms and government and by large changes in the quantity

and type of education taken up by young people. Our estimates of the edu-

cational attainment of different cohorts of young people in different macro

environments allows us to reach some conclusions about the human capital

1The Russia Longitudinal Monitoring survey, RLMS-HSE is conducted by the NationalResearch University "Higher School of Economics" and the ZAO “Demoscope,”togetherwith the Carolina Population Center, University of North Carolina at Chapel Hill and theInstitute of Sociology RAS. For more details see http://www.cpc.unc.edu/projects/rlms-hse and http://www.hse.ru/org/hse/rlms.

2

cost of the transition to a market economy. We find a substantial impact of

recession on the demand for higher education, which can potentially give rise

to large long-term costs associated with the transition.

Previous work with the RLMS has been mainly concerned with the im-

pact of transition on the structure of employment and the returns to human

capital. Berger et al. (2001) find limited training outside the worker’s own

field when compared with the United States, which, consistent with our find-

ings, would limit income mobility (see also Clerk, 2003)). Gorodnichenko

and Sabirianova Peter (2005) estimate the returns to schooling at different

quantiles of the wage distribution and find substantial lower returns at the

top of the distribution. Maltzeva (2009) finds that the return to job mobility

is small for workers over 25, so job mobility is not a source of higher returns

to education. These claims are consistent with the finding in this paper that

high earners pass the ability to earn high wages to their children independ-

ently of human capital acquisition or progression up the jobs ladder.2

We use the RLMS for the maximum currently available period, 1994-2013,

to extract earnings and other characteristics data for fathers or mothers and

their children. We then estimate the factors that explain the educational at-

tainment and earnings level of children in terms of a variety of characteristics,

including parents’education and earnings. Section 2 discusses our empirical

modelling approach. Section 3 describes the data and explains how we extract

the sample of parents and children that we use in our estimation. Section

4 reports and discusses the estimation results. Section 5 makes use of the

estimation results to derive transition probabilities across educational groups

for men and women and finally, the concluding section, gathers together the

most important findings and draws out some policy conclusions. Detailed

reporting of estimation results is collected in several Appendix tables.

2For more studies on income mobility and the returns to human capital see Lukiyanovaand Oshchepkov (2012), Gimpelson and Kapeliushnikov (2007) and Lukiyanova (2010).

3

2. Modelling approach

Following Becker and Tomes (1979), we assume that a person’s human

capital depends on investments made by parents. The rate of return on hu-

man capital (earnings) depends on family background, the person’s ability,

and luck. Family background influences the return to human capital through

a variety of mechanisms, including information and ability to take advantage

of educational opportunities, norms within the family that influence motiva-

tion and the work rate, social networks and wealth. The parents’willingness

to invest in their children’s human capital and the common family back-

ground lead to a correlation between the parents’permanent earnings and

their children’s permanent earnings, measured here by the “intergenerational

earnings elasticity”:

lnY Ci = a0 + a1 lnYPi +AXi + ei1, (1)

where lnY Ci and lnY Pi respectively are the logs of permanent earnings of

children and parents, a1 is the intergenerational earnings elasticity, Xi is a

vector of environmental characteristics, such as region and the demographics

of the local area,A is a vector of coeffi cients and ei1 is an error term reflecting

the effect of other factors such as luck.

Permanent earnings are the most appropriate measure of earning capa-

city, because current earnings depend on such time-varying factors as age

and hours worked that contaminate the connection between parents’ and

children’s earnings capacity. Regressing current earnings of children and

parents on time-varying factors and using the residuals as an estimate of

permanent earnings is a straightforward approach to obtaining permanent

earnings from observed current earnings (see, e.g., Dearden et al., 1997).

Becker and Tomes (1979) interpret the elasticity a1 as the propensity

of parents to invest in their children’s human capital. But earnings are in-

fluenced by a large number of factors beyond formal investments in human

capital. A parent with high earnings might be more likely to invest in his or

her children’s education but a parent might also have access to professional

or social networks that automatically become available to the children. We

4

distinguish between the two, formal education and unobserved other influ-

ences on earnings, by introducing the educational attainments of parents and

children in the regression.

Other omitted variables that can bias the estimate of a1 are personal

characteristics that are correlated with both parents’and children’s earnings.

We generalize our regression estimate by expanding the set of independent

variables with a number of personal characteristics to account for as many

influences on children’s earnings as possible. The coeffi cient a1 picks up

connections between parents’and children’s earnings that are unaccounted

for by the usual set of explanatory variables in earnings regressions.

We make two other adjustments to reduce measurement errors, which,

as Solon (1992) noted, bias the OLS estimate of a1 downward. First, we

average earnings over a five-year period to get rid of any impact from tem-

porary shocks that are not completely netted out by our estimation proced-

ure. Second, even when adjusted for the effect of age, earnings of younger

individuals who are at the beginning of their careers are a poor proxy for

their lifetime earning capacity. We drop very young respondents and include

only individuals with some work experience in our sample.

In light of this discussion the estimated equation is the following modified

form of equation (1):

lnY Ci = b0 + b1 lnYPi + b2E

Pi + b3E

Ci +BZi + ei2 (2)

where EPi and ECi are the educational attainment of a parent and child re-

spectively and Zi is a vector of variables that includes environmental variables

and demographic characteristics of children, such as ethnic origin, marital

status and gender. In equation (2), the investment of parents in their chil-

dren’s education is captured by the children’s educational attainment, ECi .

The demographic and personal characteristics in Zi capture other observed

influences on earnings that are unrelated to parents’ status, so the coeffi -

cient b1 on parents’earnings now captures the effect of unobserved variables

correlated with parents’earnings, the main ones being the family social and

professional networks built up by parents due to their income status. Similar

5

influences that are due to the parents’educational attainment are picked up

by the estimated b2, isolating as much as possible the impact of earnings-

related influences in b1.

We exclude children’s occupation from equation (2) because of data limit-

ations with our sample, as the sample size that includes both parents’income

and children’s occupation is very small. Since the main problem with the ex-

clusion of own occupation is that the return on own education might be

biased, we estimate also an equation that includes own occupation but drops

parents’income in order to increase sample size,

lnY Ci = c0 + c1EPi + c2E

Ci + c3OCCi +CZi + ei3 (3)

where OCCi is children’s occupation. The objective is to compare the estim-

ated returns to education in equations (2) and (3).

In addition to the equations for the transmission of earning capacity, we

estimate the following equations for the intergenerational transmission of

educational attainment

ECi = d0 + d1EPi +DQi + ei4, (4)

ECi = f0 + f1EPi + f2 lnY

Pi + FQi + ei5. (5)

where Qi is a vector of variables that includes those in Zi and some others

(the birth cohort, the total number of family’s members and the number

of children aged 3 or below and between 3 and 17, that might influence

the educational investment of parents in their children). The birth cohort

is included to capture the effect of general macroeconomic conditions on

the parents’decision to invest in their children’s education. The parents’

financial capacity to invest in their children’s education is captured by the

coeffi cient f2, whereas the coeffi cient on parents’educational attainment, d1or f1, captures the effect of family values, background and access to inform-

ation that influence children’s educational attainment independently of how

much command the family has over financial resources. We estimate separ-

ately equations (4) and (5) because of limitations in our sample. The sample

6

that does not include parents’income is much larger.

The full effect of parents’educational attainment and income on children’s

earnings is obtained when ECi is substituted from equation (5) into equation

(2) to obtain the coeffi cient b2 + b3f1 on parents’education and b1 + b3f2 on

parents’permanent income.

Estimates in the literature show a positive and significant coeffi cient of

intergenerational transmission of permanent earnings, a1, practically in all

countries for which evidence is available (see Solon, 1999; D’Addio, 2007).Cross-country comparison of intergenerational income elasticities based on

various measures of earnings for pairs of fathers and sons suggests that the

transmission is highest in the United Kingdom, Italy, the United States and

France. In these countries at least 40% of the economic advantage that

high-earning parents have over low-earning parents is passed on to the next

generation. In contrast, intergenerational transmission is comparatively low

in the Nordic countries, Australia and Canada, with less than 20% of the

earnings advantage passed from parent to offspring.

National peculiarities as to the financing of education, the degree of open-

ness of the labour market and parents’preferences for investing in the human

capital of the next generation are factors cited for the international differ-

ences in the intergenerational income elasticity. Solon (1999) describes a

society with a very high degree of transmission as a “cast” society, where

children inherit their parents’position in the wage distribution.

Empirical estimates of equations (2) and (5) for several countries show

that parents’education mainly influences children’s earnings through educa-

tional attainment (i.e., via b3f1). But in some countries (the United King-

dom, Spain, Italy, Netherlands, Luxembourg and Ireland) the direct linkage

via b2 is more important. In most countries the direct effect of father’s edu-

cation, measured by b2, is not statistically significant in determining their

children’s wages, once the children’s educational attainment is taken into ac-

count (Causa et al., 2009). As for earnings transmission, there is evidence

that there is both direct transmission, through the coeffi cient b1, and indirect,

through the impact of parents’income on the child’s educational attainment,

measured by the coeffi cient b3f2 (D’Addio, 2007, Chevalier et al., 2013)

7

When one estimates equations (1) and (2) by OLS on the entire sample,

the effect of explanatory variables is identical across the entire distribution

of the dependent variable. Becker and Tomes (1979) argue that low-earning

parents are constrained in their ability to finance the education of their chil-

dren, whose earnings as a result are below the earnings of children with the

same ability from unconstrained families. Following this argument, Bratberg

et al. (2005) used a quantile regression to explore the idea that the impact

of the explanatory variables is different at different ranges of the distribu-

tion of the dependent variable (the log of permanent earnings). We test this

hypothesis by also estimating quantile regressions.

3. Data and choice of variables

The Russia Longitudinal Monitoring Survey (RLMS) is the primary source

of data used in this paper. The RLMS is an annual multi-wave data set of a

representative sample of more than 10,000 individuals that started in 1992.

Because the survey was redesigned in 1994 and lost consistency with the

previous two waves, we use data for the 18 years 1994-2013, 2013 being the

latest available year at the time of writing.

The RLMS started as an unbalanced panel of 4,000 household addresses.

New households are added in two different ways. First, children who leave

the parents’original address are followed up and if they remain in the same

local area their new address is added to the survey. If they leave the local

area they are dropped from the survey. Second, a family that moves within

the same local area is traced and its new address is added to the survey;

the original household address is not removed from the survey and the new

resident is added to the survey. Attrition is due to natural causes, refusal

to continue participation and moving to another area, as no effort is made

to trace respondents who have left the original locality. On average 8% of

respondents were lost each year from 1994 to 2009, before the sample size

changed more fundamentally in 2010. In that year the sample was increased

by about 2,000 addresses, in order to improve its representativeness across

the regions of the country. The sampled population increased from 13,991

individuals in 2009 to 21,343 in 2010.

8

Parents’ and children’s personal characteristics can be matched from

household data when children still belong to the household, whereas income

and educational attainments are matched by combining parents’responses

with those of children who have left the household but remained within the

same local area. When children leave home and move outside the local area

the link is lost, as those children are dropped from the survey. Although this

feature of our sample probably puts more weight on children who are born

in the large cities, we have enough observations from outside the cities to

obtain reliable estimates.

There is a large number of useable characteristics in the survey and a

number of different measures of income. Income measures include wages

and salaries at a primary place of work, earnings elsewhere, wages owed by

employers but not yet paid, earnings in-kind and total income from all sources

(including e.g., income from capital). We consider wages and salaries at the

primary place of work to be the most appropriate measure for our research,

as we are looking for estimates of permanent income (see also the discussion

below in the context of the estimated equations). In line with this decision,

we also use the reported hours of work in the primary place of work as our

measure of working time.

The survey includes information on years of schooling and academic de-

grees obtained. Educational standards in the Soviet Union and then in Russia

have changed several times, so to obtain the same degree in different periods

often required a different number of years of schooling. In order to achieve

consistency of the measures of education across individuals in different years

of the survey we use a categorical variable indicating the highest academic de-

gree obtained. The most appropriate classification given the size and nature

of our sample is a three-category measure of educational attainment, as fol-

lows: secondary school or below, technical school (full secondary education

plus some professional training or attendance at a technical trade school,

which includes technical, medical, music, pedagogical or art school) and fi-

nally tertiary (university or equivalent) education. More detailed breakdowns

gave either too small samples or inconsistent results due to memory lapses

in older individuals.

9

The indicators for occupation are created according to the first digit of

the ISCO-88 codes, but because some related occupational groups ended up

with too small a number of respondents we aggregated further. We merged

offi cials, managers and professionals; clerks and service workers; and skilled

agricultural and fishery workers and unskilled workers, guided mainly by the

fact that average wages within these three categories of workers were similar

across the group.

A peculiarity of the Russian economy is the considerable heterogeneity of

the distribution of human capital across geographic regions and settlements

of a different size. Economic activity in Russia is concentrated in places where

the state authorities are situated, as well as in regions associated with the

extraction and processing of minerals and the transportation of the extracted

products. In this regard, the Moscow region is of particular importance

for the economy. The Russian centres of economic activity often coincide

with centres of professional training. All of the above in combination with

low internal labour mobility cause the large geographical variation of labour

demand and supply. Indicators for region and the type of the settlement

where a respondent is resident are used to proxy differences in the economic

activity and resulting real wages in different local labour markets. There

is a wealth of regional data and eight regions are distinguished: Central;

North-Western; Volga; Southern; North-Caucasian; Ural; Siberian; and Far

Eastern.

Table 1. Rules of the assignment of observations to settlement types

Category Population per locality

Capital cities 4.5 mln. or more

Big cities 0.8-1.5 mln.

Medium-sized cities 180-800 thousands

Small towns 30-130 thousands

The countryside less than 30 thousands

Assigning observations to settlement types is generally guided by popu-

lation size as shown in Table 1. In addition to these rules some special rules

10

are applied to assign settlements located in the Moscow administrative re-

gion. Due to the fact that these localities are close to the city of Moscow, the

demand for labour from outside the locality is generally high, so that average

earnings in localities in the Moscow region are almost as high as they are in

Moscow. But residents of the city of Moscow and of the settlements located

in the Moscow region are very different in terms of educational attainments.

Average educational attainment in the Moscow region is considerably lower

than it is in the city and comparable to that of persons living in other non-

capital settlements of a corresponding size. We interpret these differences as

resulting from mobility of labour between Moscow and its region but immob-

ility of school children who attend school according to habits and customs in

their local area. Consequently, we reclassified settlements located in the Mo-

scow region as capital cities in equations (1)-(3), to avoid biasing the results

of the smaller categories.

A person is defined as married if he or she cohabits with another, irre-

spective of whether the marriage is offi cially registered or not. Apart from

preferring this definition because offi cial registration is sometimes connected

with religion or other factors that are not relevant to our study, this defini-

tion of marriage is also necessitated by the survey questionnaire, which did

not distinguish in all years whether a marriage is registered or not.

Given our interest in estimating the impact of informal networks on the

transmission of earnings potential from parents to children, ethnic origin

can be an important influence, with non-Russians placed at a disadvantage.

The RLMS data set reports several ethnic groups or nationalities living in

Russia and we aggregate them into four groups. The first is the core group

of Russians, Ukrainians and Belorussians. The second group encompasses

Christian/European groups not included in the first group, whose origins are

mainly Polish, Spanish, Armenian and some others. The third group includes

mainly Islamic people such as the Tartars and the Muslims of Central Asia

and North Caucasus. Other nationalities such as the Buryats, the Chuvash

and the Kalmyks are aggregated into the fourth group.

Six periods encompassing children born in 1947-55, 1956-60, 1961-65,

1966-70, 1971-75 and 1976-80 are identified and cohort (time) dummies are

11

included in all regressions. Children born in 1981-83 are the default group.

Because older workers of both genders are under-represented in the samples

and equation (5) is estimated with a smaller sample size, for this equation

we aggregate the cohorts into three, one for the older ones born in 1947-75,

one for the younger cohorts born in 1976-1980 and the reference group born

in 1981-83.

Various measures of household characteristics are included in our regres-

sions. The definition of a household in the RLMS includes people living

together and having common income and expenditures, including unmarried

children under 18 who study in a different population centre. The total num-

ber of family members, the number of children 17 or under and the number

of children 3 or under are used as measures of household composition. The

variable for residence ownership takes 1 for individuals living in their own

dwelling and 0 for those occupying a rented property or living in a dormitory.

Car ownership takes value 1 for all members of the household if there is at

least one passenger car at the disposal of the household, and 0 otherwise.

The living-space square of dwelling specifies the number of square meters of

living-space at the disposal of the household.

Several time series published by the state statistical agency Rosstat are

used in our estimation. Nominal income measures by region are adjusted by

the cost of a consumer basket by region to arrive at real regional incomes.

Any variable entering equations (1)-(5) is defined as the average for an indi-

vidual taken over all non-missing values in all rounds of the survey in which

an individual participated. The averages of binary dummy variables are then

rounded to the nearest integer to keep the binary format. The averages are

estimated on the sample of individuals of age 30-65 to exclude those who are

at the beginning of their career, whose income is subject to measurement

error.

Because of data availability, we used slightly different samples across the

estimated equations, aiming always to get the maximum information in each.

Appendix Table 5 shows all the descriptive statistics of our samples. Equa-

tions (1) and (2) were estimated with the sample shown in column 1 of Table

5. It includes all parent-child pairs, of either gender, who reported earnings

12

at the primary place of work in at least five rounds of the survey (and have

at least one observation on the other variables included in the regressions).

If both parents report earnings and education the parent with the higher

income is used.

Equation (3) consists of the same variables as equation (2) except for

the omission of parents’earnings and the inclusion of occupational dummies.

The equation may contain parents who do not have income data and were

therefore not present in the estimation of equation (2). In this case the edu-

cation of the father is used or, if not available, that of the mother. Equation

(4) is estimated on two different samples. The first sample is in column 3

of Table 5 and is for sons and fathers. The second, in column 4, estimates

equations for daughters and fathers. Equation (5) repeats the estimation of

equation (4) but also includes the father’s earnings.

The variations in the mean values reported in Table 5 are not large but

some comments are in order about the earnings variables and their implica-

tions for other variables. The mean year of birth is especially sensitive to the

presence of earnings in the specification, given our way of averaging earn-

ings over five years. When earnings are in the equation the sample includes

only individuals who have reported earnings in at least five rounds of the

survey. As the probability that young people report earnings increases with

age, whereas for older people it decreases with age, the inclusion of earnings

of children raises the mean age, whereas the inclusion of parents’earnings

lowers the mean age. This explains the differences in mean age across the

columns in Table 5.

Finally we need to point out a caveat about the RLMS survey, which does

not chase individuals who change locality. If family networks and labour

market influences are stronger in the family’s local area, persons who move

out should expect a diminished impact of parents’networks on their own

income. As internal migration in Russia is low, however, this should not

introduce a serious bias in our estimates but it should be noted that our

estimates are for parents and children who live in the same locality.

13

4. Econometric specification and estimation results

In order to estimate our model we first need to assign a good measure of

permanent earnings to parents and children. As we already noted, permanent

earnings are obtained as the residuals of a regression of current real monthly

earnings on a vector of characteristics that remove temporary influences. The

objective is to obtain an underlying rate of return to human capital for each

person.

Age and hours worked are two obvious variables whose influence on the

rate of return to human capital should be netted out. A more diffi cult issue

arises because of the transition of Russia from a planned economy to a market

economy in the 1990s. Human capital acquired before the transition was

directed to different activities from the ones available in the market economy

of later years, and so was less productive than human capital acquired after

the transition. In order to correct for this change we introduce the birth year

in the earnings regression, and refer to it as a cohort effect. Any other cohort

effects may also be picked up by this procedure.

General macroeconomic effects are normally netted out by introducing

year dummies, but having introduced age and cohort effects we would not

be able to identify the year effects. One possibility is to use average earnings

in the RLMS in place of year dummies to pick up macro effects, but because

the RLMS is not a balanced survey over incomes this may not be a good

proxy. The RLMS was not designed to represent the working population and

although the moments of the distribution of earnings reported in the RLMS

are close to the ones available in the Rosstat national series, some differences

exist. In particular, the sample mean of earnings varies from period to period

due to attrition, the addition of new households, movements in and out of

unemployment, changing occupations and industries by some workers, and

so on. For this reason we introduce extraneous information obtained from

Rosstat, in the form of the real wage index, which is balanced over the

whole economy (see also Borisov, 2007). The real wage index is the ratio of

average real wages in period t to average real wages in some initial period

0 : RWIt = ARWt/ARW0. The earnings function with the logarithm of the

14

real wages index estimated on the repeated cross-sections is:

lnRWit = k0 + k1 lnRWIt +KVit + eit6 (6)

where lnRWit is the logarithm of the individual’s real monthly earnings and

Vit is the vector of the other variables that we discussed. Since lnRWIt =

lnARWt − lnARW0 and lnARW0 is independent of time and person, equa-

tion (6) gives

lnRWit = (k0 − k1 lnARW0) + k1 lnARWt +KVit + eit6 (7)

To the extent that the logarithm of average real wages in period t, lnARWt,

corresponds to the true underlying period mean of the dependent variable

it picks up the average effect of macroeconomic factors on real earnings in

period t; it proxies for period, being unrelated with age and cohort. The

results reported by Borisov (2007) confirm that lnRWIt is a good proxy for

period.

Equation (6) is our preferred equation from which we obtain permanent

earnings, after we correct for possible selection bias. Selection bias may arise

for two reasons. First, if people’s decisions about working or not working de-

pend on earnings, people with low earning capacity will be under-represented

in the sample. Second, reporting earnings may be dependent on the level

of earnings. Aivazjan and Kolenikov (2001) show that the probability of

responding is a negative function of household’s expenditures and so house-

holds with higher earnings are less likely to report. Selection bias because

of the participation decision is likely to be relevant only for women, but we

cannot estimate equation (6) on the female sample, because Rosstat reports

the real wages index for all working individuals, not males and females separ-

ately. Correction for the entire sample together is also not possible because

of the absence of hours information for non-participants. Had the elasticity

of earnings with respect to hours been 1, or close to it, we could have avoided

this problem by dividing earnings by the number of hours of work to arrive

at hourly wages but results reported in Appendix Table 6 indicate that the

15

elasticity of earnings with respect to hours worked is in the range 0.3-0.4.

Any selection bias from failure to report, however, can easily be accounted

for. The earnings equation becomes:

lnRWit = g0 +∑n

g1nDBYni +∑m

g2mDAmit + g3 lnRWIt

+ g4 lnHRSit + eit7 (8)

with the selection equation,

Selit = h0 + h1MARSTit + h2OWNHSit + h3SQLIV Eit + h4OWNCARit

+ h5NCHLD3it + h6AGEit + h7AGESQit + eit8. (9)

In this equation the indices i, n,m and t stand for individual, birth year, age

and period respectively; lnRWit is the logarithm of real monthly earnings

from the primary place of work; DBYni are dummy variables for birth year

defined as DBYni = 1 if the birth year is the year of observation minus the

respondent’s age plus one and DBYni = 0 otherwise; DAmit are dummies for

age with the one-year window: DAmit = 1 if the age of the respondent is the

year of observation less the year of birth plus one and DAmit = 0 otherwise;

lnRWIt is the logarithm of the real wages index; lnHRSit is the logarithm of

working hours; Selit is a dummy for selection, which takes value 1 for working

individuals who reported earnings and 0 for working individuals who refused

to declare earnings from the primary place of work; MARSTit is a dummy

for marital status; OWNHSit is a dummy for residence; SQLIV Eit are the

square meters of a living space; OWNCARit is a dummy for car ownership;

NCHLD3it is the number of children aged three or less; AGEit is years of

age and AGESQit the same squared.

Equations (8) and (9) are estimated on the repeated cross-sections, in-

cluding observations for all the respondents from all the rounds of the RLMS

survey. The residuals from equation (8) are the measure of the permanent in-

comes that we use in equations (1)-(3), and (5). As an individual is normally

represented by several observations in equations (8)-(9) and all children are

16

selected in equations (1)-(5), some individuals and parents appear more than

once in the estimated equations. We use the robust estimator that adjusts

the standard errors to allow for the intragroup correlation within the clusters

of individuals or parents. Equations (1) and (2) are estimated by OLS and

quantile regressions. Equation (3) is estimated by OLS only.

As stated in Section 3, the parents in equations (1)-(3) are selected on the

basis of gender and certain income characteristics. To ensure the consistency

of estimates, equations (1) and (2) were augmented by two dummies, the first

of which is parent’s gender, and the second taking the value 1 if the parent

is chosen because he or she earns more than the other parent.

The dummy variable of occupation entering equation (3) is constructed as

an individual mean averaged over periods. The averaging of a time-varying

characteristic increases the error-in-measurement, which causes the down-

ward bias of the estimate. Therefore, one may expect the estimates of the

effect of occupation from equation (3) to be lower bounds of the true effects.

In equations (4) and (5) the dependent variable is a categorical one indic-

ating the level of children’s education as one of three levels, so the ordered

response model is used to estimate the effect of parent’s educational attain-

ments on their children’s education. As the distribution of residuals from

equations (4) and (5) appear to be closer to a normal distribution than to a

logistic one, the ordered probit model is used.

The earnings-age profile is estimated by the Heckman selection procedure

according to equations (8) and (9). The sample includes the data from all

the waves of the survey on males and females aged 25 to 65 years who re-

ported earnings and hours at the primary place of work. Selected estimates

are presented in Appendix Table 6. The correlation coeffi cient between the

wage and selection equations is statistically significant. The estimates of the

selection equation have predictable signs. The selection equation confirms

that individuals with more assets are less likely to declare earnings.

The earnings-age profile is convex upward having the expected shape,

with a peak at age 45.5. The cohort effect is maximum for those born in

1983. The 1983 cohort entered the labour market mainly in the first half of

the 2000s, when the demand for labour grew rapidly after the financial crisis

17

of the late 1990s. As we would expect given the structural transformation of

the early 1990s, the cohort effect decreases gradually as we move from 1983

back; the results indicate that the human capital of older people is valued

less than that of younger ones.

The estimate of the logarithm of the real wage index is 0.87 and strongly

significantly different from both 0 and 1. The high value of the estimate

confirms that the index is a good measure of the average real wages of our

sample, but the fact that it is less than 1 shows that the wages of respondents

in our sample are less cyclical than the average. The estimate of the logarithm

of monthly hours is low, at 0.35, confirming our claim that dividing monthly

earnings by hours worked is not a good way of obtaining hourly wages as

the rate of return to human capital (see also Maltzeva, 2009, for a similar

conclusion).

The residuals from equation (8) are used as the measure of the permanent

income of children and parents in equations (1)-(3) and (5).

Table 7 presents the OLS estimates of equation (1). The earnings elasti-

city is about 0.33 and strongly significant. Equation (1) is also estimated for

the 20th, 40th, 60th, and 80th quantiles of the distribution of its dependent

variable by means of the quantile regression. The estimates of the effect of

parent’s earnings for the specified quantiles are shown in Table 2.

Table 2. The estimates of the earnings elasticity for quantiles

Quantiles

0.20 0.40 0.60 0.80

The earnings

elasticity

0.2933***

[0.0947]

0.3537***

[0.0738]

0.3744***

[0.0709]

0.4695***

[0.0893]

*** Significant at .01 level.

Bootstrap standard errors are in brackets.

The results of the quantile regressions demonstrate heterogeneities in the

estimates, with the earnings elasticity increasing as we move up the (chil-

dren’s) wage distribution. But the Wald test of the hypothesis that the

earnings elasticities estimated for the 20th and 80th quantiles are equal to

18

each other cannot reject equality at the 10% level. So although the change in

the point estimate is evident and monotonic, it is not statistically significant.

The OLS estimates of equation (2) are in Appendix Table 8. The elasti-

city with respect to parent’s permanent earnings is 0.25 and since education

variables are also in the regression, it picks up the intergenerational earnings

elasticity that is due to the family’s social networks. If we take the 0.33

estimate from equation (1) as the total effect, we conclude that most of the

transfer, about three-quarters, is due to networks and the remainder quarter

to educational transfers.

As expected, own education is a significant determinant of earnings.

Technical school graduates receive a 14% wage premium, whereas university

graduates gain an extra 35% increase in their permanent income when com-

pared with the default group of no education beyond secondary level. The

effect of parent’s education on children’s earnings is not significantly different

from 0. It is possible that the endogeneity of the independent variables in

equation (2) masks the full effect of parent’s education. Parent’s education

influences their children’s educational attainment and also determines par-

ent’s earnings. But the fact that once these indirect effects are controlled for

there is no significant effect of parent’s education remaining, indicates that

highly-paid parents, but not necessarily highly-educated ones, provide their

children with beneficial social connections. This is also valid if households

with highly-educated parents cultivate a stronger work ethic: such an ethic

is not necessarily accompanied by a higher monetary return.

Equation (2) is also estimated for the 20th, 40th, 60th, and 80th quantiles

of the distribution of children’s earnings by means of the quantile regression.

The selected estimates for the specified quantiles are shown in Appendix

Table 9. The estimates do not reveal any consistent pattern across quantiles

(note that the sample for some of these quantiles is very small).

The estimates of equation (3) are in Appendix Table 10. Comparing the

estimates of equations (2) and (3) reveals that the inclusion of occupational

dummies does not change the effect of the educational attainment of children.

Other results show that the effects of ethnic background are close to 0 and

statistically insignificant for all ethnic groups. The type of the residential

19

settlement, however, affects long-term earnings. The difference in earnings

between those who live in the metropolitan area and in the countryside is

statistically significant at the 1% level. As expected, wages are positively

correlated with the size of settlement. People living in rural areas on aver-

age earn 46% less than those with similar qualifications in the metropolitan

areas. Region plays a less important role, with only the North Western

Federal region associated with higher wage earnings, most likely because of

compensation for poor climatic conditions. The occupational differences are

also as expected, with clerks and service workers, skilled agricultural and fish-

ery workers and the unskilled earn significantly less than offi cials, managers,

and professionals, even when educational attainment is taken into account.

The gender difference in earnings is large and strongly significant, at a 41%

premium of male over female workers.

Equation (4) estimates the transmission of educational attainment from

parents to children. The results of the ordered probit for the sample of sons

and fathers are presented in Appendix Table 11 and those for daughters and

fathers in Appendix Table 12. Examination of the results shows that father’s

education attainment strongly affects educational achievements of both sons

and daughters, with very similar point estimates. Other estimates show

that Muslim children, regardless of gender, are less likely to graduate from a

technical school or university than the reference group consisting of Russian,

Ukrainian, and Belorussian. Compared to the Central Federal region which is

the reference one, the expected level of education of both sons and daughters

is significantly lower in the Urals and Siberian Federal regions. The latter

are regions that specialize in the extraction and processing of raw materials,

which require less highly-educated labour.

Educational attainment also depends on year of birth. The probability

of sons and daughters born in 1981-83 to achieve a higher level of education

is significantly larger than that of other children. Children from the 1981-

83 cohort made the decision about education in the late 1990s and early

2000s, when the demand for highly-educated workers was increasing fast.

The difference in the acquisition of higher education is particularly high when

we compare children born in 1981-83 with those born in 1971-1975. The

20

latter cohort completed secondary school in late 1980s and early 1990’s, when

the transition to the market economy started and there was a deep drop in

the demand for skilled labour. The impact of such differences on earnings

and productivity is long-lasting and shows how temporary falls in economic

activity can have long-term effects on the economy.

The composition of the family has an impact on the educational attain-

ment of daughters, in that daughters in larger households do not acquire as

much higher education as daughters in other households. In larger house-

holds daughters are more likely to drop out of formal education after second-

ary school and, one assumes, devote more time to household work such as

childcare or elderly care.

Equation (5) adds parent’s permanent earnings to the explanatory vari-

ables of equation (4). The estimation results are shown in Appendix Table

13. The results show that once the educational attainment of the father is

controlled for, father’s income does not play a role in the educational invest-

ments in the children (male or female). This shows both a strong intergen-

erational linkage in educational attainments but also the absence of financial

constraints for children in higher education, which would have made parent’s

income a more important factor in their education decision. We note that

in the USSR education was completely free and even after the transition a

significant number of college and university students had been attending on a

non-tuition basis. According to the Russian Statistical Yearbook 2014, only

one third of all the university students paid for education in 2000, though

this share increased to 50% by 2005 and it remained about 55-60% in sub-

sequent years. As the samples estimated in this paper cover respondents born

before 1984, most children had graduated when education was offered on a

non-tuition basis. The result may not hold for later cohorts who attended

after tuition became more widespread. The results estimated in this context

are also consistent with our earlier finding, in equations (1) and (2), that the

main channel of transmission of earnings capacity from parents to children

is a direct one that utilizes other channels and not one that works through

the educational investments of parents in their children.

21

5. Intergenerational educational mobility

The results of the estimation of the ordered probit model can be used to

predict the probabilities of achieving each level of educational qualification

and compute the intergenerational mobility probabilities across educational

groups. Equation (4) is used for that purpose, because its bigger sample

gives more accurate and consistent estimates than equation (5). Table 3

displays the predicted probabilities of achieving the levels of education for a

child of the Russian, Ukrainian or Belorussian nationality, born in 1971-1975,

who is resident in the Volga region in the countryside and whose household’s

composition corresponds to the means of the respective samples.

Table 3. The predicted conditional probabilities of achieving educationattainments, the rural areas of the Volga region, cohort born in 1971-1975.

Son’s level of education Daughter’s level of education

Father’s level of

education

I II III I II III

I 0.745 0.223 0.032 0.511 0.388 0.101

II 0.551 0.355 0.094 0.323 0.461 0.216

III 0.385 0.430 0.185 0.174 0.447 0.379

I - secondary or below secondary education; II - technical school; III -

university or higher.

Table 3 demonstrates high intergenerational persistence in educational

qualifications for lower grades, as shown by the values of the diagonal terms.

However, it also shows mobility, especially for men. Generally, upward mo-

bility is more prevalent for daughters, as shown by the higher entries in the

cells to the right of the diagonal, than it is for sons, who are characterized

by more downward mobility.

Consider first mobility across adjacent categories. Perhaps surprisingly,

Table 3 shows that sons with fathers who have beyond secondary education

are likely to be less well educated than their fathers, and the same holds for

daughters but only if the father has a university degree. Next consider long-

distance mobility, defined as the probability of a person from the bottom or

22

top tail of the distribution to end up in the opposite tail. Bottom-to-top edu-

cation mobility is measured as the probability of the child achieving tertiary

education conditional on the father having secondary or below-secondary

education. Conversely, top-to-bottom education mobility is measured as the

probability of the child belonging to the low-educated category conditional

on the father having tertiary education.

It follows from Table 3 that the bottom-to-top probabilities for sons is

0.032 and for daughters 0.101, whereas the top-to-bottom probabilities are

0.385 and 0.174 respectively. The downward mobility for males may seem

surprising but one can argue that it is peculiar to the 1971-1975 cohort,

which made their educational decisions at the time of the transition. The

drop in demand for skilled labour was large at the time, especially outside the

metropolitan areas. The probabilities improve somewhat when the Moscow

and St. Petersburg metropolitan areas are considered instead of the Volga

one, with the bottom-to-top mobility for sons increasing to 0.110 and the

top-to-bottom decreasing to 0.180. For daughters the bottom-to-top mobility

increases to 0.329 and the top-to-bottom one falls to 0.038.

In order to test further the hypothesis that educational mobility depends

on the demand conditions for skilled labour, we also computed the transition

probabilities for children of the 1981-1983 cohort, who entered education in

the boom years around the new millennium. Offi cial figures provided by the

State Committee of the Russian Federation on Statistics (2002) and Rosstat

(2014) show that total employment increased by 6.73 million persons in 2001-

2013, whereas employment of offi cials, managers, and professionals increased

by 6.65 million, so almost the entire increase in total employment between

2001 and 2013 can be accounted for by the increase in the employment of

high-skilled labour. The estimates of equation (4) give the transition prob-

abilities in Table 4, which is directly comparable to Table 3.

23

Table 4. The predicted conditional probabilities of achieving educationattainments, the rural areas of the Volga region, cohort born in 1981-1983.

Son’s level of education Daughter’s level of education

Father’s level of

education

I II III I II III

I 0.575 0.341 0.084 0.375 0.448 0.177

II 0.367 0.435 0.198 0.210 0.460 0.330

III 0.223 0.443 0.334 0.099 0.385 0.516

I - secondary or below secondary education; II - technical school; III -

university or higher.

Table 4 shows that as before, upward mobility is more likely for females

than males, who still exhibit at least as much downward mobility as up-

ward. However, transitions generally improve. The bottom-to-top transition

probabilities for the 1981-83 cohort of sons and daughters increases by up

to 0.084 and 0.177 respectively, or by 2.6 and 1.75 times, as compared with

the corresponding figures for the 1971-75 cohort. The top-to-bottom trans-

itions for the 1981-83 cohort of sons and daughters are 0.223 and 0.099, both

more than 40% lower than the equivalent ones of the 1971-75 cohort. The

intergenerational transmission of educational attainment is even more pro-

nounced for children living in the metropolitan areas, with the bottom-to-top

transition for sons increasing to 0.224 and the top-to-bottom decreasing to

0.083. For daughters the bottom-to-top transition increases to 0.462 and the

top-to-bottom one falls to 0.017.

6. Conclusions and policy implications

In this paper we estimated the intergenerational transition of earnings in

post-transition Russia, which takes place either through educational quali-

fications that correlate across generations or through other channels, such as

family and professional networks. Our estimation results, with a longitud-

inal data set that extends from 1994 to 2013, show that the intergenerational

transmission of earnings is high and, importantly, the channels through which

24

it is manifested are the informal networks not picked up by formal education.

In other words, high-income parents do not necessarily invest more in their

children’s education, but despite this, their children are more likely to end up

being higher earning individuals than the children of poorer families. The in-

tergenerational earnings elasticity is estimated to be equal to 0.33 on average,

and although it rises as incomes rise, the estimated difference for the 20%

and 80% quantiles of the earnings distribution is not statistically significant.

Other countries with similar magnitudes of the international earnings elasti-

city include Italy, the United States and France. In these countries, however,

the high values of the intergenerational earnings elasticity is broadly attrib-

uted to private investments in children’s human capital resulting from lower

public expenditures on education (Black and Devereux, 2011). The Russian

case seems different, because a relatively large value of the elasticity is com-

bined with tuition-free education prevailing among the respondents in our

sample.

The high intergenerational earnings elasticity implies low earnings mo-

bility, so the mechanisms behind it can be considered to be obstacles to the

improvement in the welfare of people at the bottom of the earnings distribu-

tion. The fact that the causes of the high elasticity are not related to access

to education makes it more diffi cult for policy to break this barrier because

it is not obvious where it should be targeted. For example, the strong in-

tergenerational income transition could be due to closed professions whereby

sons and daughters enter the father’s or mother’s profession more easily than

those from outside could do; it could be due to social networks, where par-

ents influence their colleagues and acquaintances when their children apply

for jobs; or it could be the outcome of corruption where corrupt practices

are passed down from parents to children. It is clearly the case, however,

that a policy aimed at the enhancement of transparency and competition in

the labour market could shed more light on the correlations identified in this

paper and promote more intergenerational earnings mobility.

As in other studies we find that own education contributes to one’s per-

manent earnings, so improving one’s own education is an effective way of

improving one’s position in the income distribution, especially for children

25

from low-income families. The fact that we find that parents’income does

not influence educational attainment is indirect evidence that there are no

binding budget constraints in educational choices. This plausible finding may

be due to the fact that most of the children of our sample graduated during

a period when education in Russia was offered free of tuition. This could

change as tuition is gradually being introduced in many institutes of higher

education.

The results show that in contrast to parents’income, parents’education is

a strong influence on children’s educational attainment. So enhancing one’s

education has beneficial intergenerational effects on children’s education too.

The mechanism is not through budget constraints, but through incentives

and knowledge of educational opportunities and their contribution to career

enhancement.

An important finding of our study is that controlling for parent’s educa-

tion and income, the local labour market and the state of the macro economy

are strong influences on the acquisition of education by children. In times of

lower demand for skilled labour, fewer children enter institutes of higher edu-

cation. Of course, this is ineffi cient given the length of macro recessions and

the impact of education on lifecycle earnings and introduces cohort effects

on earnings well into adult life. Children who live in larger local markets

are more likely to pursue higher education, because of the many more op-

portunities offered for skilled labour in bigger cities. In addition to this

effect, cohort effects are estimated. Children belonging to our youngest co-

hort, born in 1981-1983, were more likely to graduate from a university than

earlier cohorts were. When the children born in 1981-83 were making their

educational choices the demand for skilled labour in Russia was the highest

in the post-transition era. Our findings point to the importance of promoting

the creation of jobs for skilled labour in depressed areas or in rural environ-

ments in order to give incentives for children from those areas to pursue more

education.

26

References

[1] Aivazjan, S. and S. Kolenikov (2001). “Poverty and Expenditure Differ-

entiation of the Russian Population”. EERC’s Working Paper Series,

01/01.

[2] Becker, G. and N. Tomes (1979). “An Equilibrium Theory of the Distri-

bution of Income and Intergenerational Mobility”. Journal of Political

Economy, 87, 1153-1189.

[3] Berger, M. C., J. S. Earle and K. Z. Sabirianova (2001). “Worker training

in a restructuring economy: Evidence from the Russian transition”, in

Solomon Polachek (ed.) Worker Wellbeing in a Changing Labor Market

(Research in Labor Economics, Volume 20) Emerald Group Publishing

Limited, 159-190.

[4] Black, S. E. and P. J. Devereux (2011). “Recent Developments in In-

tergenerational Mobility”. In: Handbook of Labor Economics, Vol. 4B.

Edited by D. Card and O. Ashenfelter. Elsevier B.V., North Holland.

[5] Borisov, G. (2007). “The Vintage Effect on the Russian Labor Market”.

Eastern European Economics, 45, 23-51.

[6] Bratberg, E., O. A. Nilsen and K. Vaage (2005). “Intergenerational Mo-

bility: Trends Across the Earnings Distribution”. University of Bergen

Department of Economics Working Papers in Economics, No. 04/05.

[7] Causa, O., S. Dantan and Å. Johansson (2009). “Intergenerational Social

Mobility in European OECD Countries”. OECD Economics Department

Working Papers, No. 709.

[8] Chevalier, A., C. Harmon, V. O’Sullivan and I. Walker (2013). “The

impact of parental income and education on the schooling of their chil-

dren”. IZA Journal of Labor Economics, 2:8.

[9] Clark, A. (2003). “Returns to Human Capital Investment in a Transition

Economy: The Case of Russia, 1994-1998”. International Journal of

Manpower, 24, 11-30.

27

[10] D’Addio, A. C. (2007). “Intergenerational Transmission of Disadvant-

age: Mobility or Immobility across Generations? A Review of the Evid-

ence for OECD Countries”. OECD Social, Employment and Migration

Working Papers, No. 52.

[11] Dearden, L., S. Machin and H. Reed (1997). “Intergenerational Mobility

in Britain”. Economic Journal, 107, 47-66.

[12] Gimpelson, V. and R. Kapeliushnikov, eds. (2007). Wages in Russia:

Their Evolution and Differentiation. Moscow: HSE Publishing House.

(in Russian)

[13] Gorodnichenko, Y. and K. Sabirianova Peter (2005). “Returns to School-

ing in Russia and Ukraine: A Semiparametric Approach to Cross-

Country Comparative Analysis”. Journal of Comparative Economics,

33, 324-350.

[14] Lukiyanova, A. (2010). “The return to education: what the meta-

analysis shows?” Higher School of Economics Economic Journal, 14,

326-348. (in Russian)

[15] Lukiyanova, A. and A. Oshchepkov (2012). “Income mobility in Russia

(2000-2005)”. Economic Systems, 36, 46-64.

[16] Maltzeva, I. (2009). “Labor mobility. What is the return on specific hu-

man capital in Russia?”Higher School of Economics Economic Journal,

243-278. (in Russian)

[17] Solon, G. (1992). “Intergenerational Income Mobility in the United

States”. American Economic Review, 82, 393-408.

[18] Solon, G. (1999). “Intergenerational Mobility in the Labor Market”.

In: Handbook of Labor Economics, Vol. 3A. Edited by D. Card and O.

Ashenfelter. Elsevier B.V., North Holland.

28

Appendix

Table 5. Descriptive statistics on estimated samples.1

Sample 1 2 3 4 5 6

Estimated

equation

(1)

and

(2)

(3) (4) (4) (5) (5)

Sample size 250 733 848 803 240 239

Parent’s

permanent

earnings2

0.018

[0.568]

0.200

[0.549]

0.231

[0.549]

Year of parents’

birth

1948.7

[5.261]

1943.0

[7.563]

1947.5

[8.237]

1947.7

[8.179]

1951.9

[5.571]

1951.7

[5.158]

Female parents 0.596 0.591 0.000 0.000 0.000 0.000

Parent’s education

below-secondary or

secondary

0.312 0.477 0.458 0.416 0.325 0.331

Parents with

technical school

0.480 0.359 0.384 0.412 0.496 0.477

Tertiary parent’s

education

0.208 0.164 0.158 0.172 0.179 0.192

Children’s

permanent

earnings2

0.020

[0.539]

-0.009

[0.553]

Year of children’s

birth

1973.5

[3.792]

1969.7

[6.052]

1974.4

[6.418]

1974.9

[6.413]

1977.3

[4.502]

1977.9

[4.070]

Female children 0.548 0.574 0.000 1.000 0.000 1.000

Capital cities’

residents30.168 0.194 0.118 0.119 0.117 0.092

Big cities’residents 0.228 0.172 0.120 0.131 0.150 0.138

Middle-sized cities’

residents30.224 0.199 0.217 0.242 0.233 0.243

29

Sample 1 2 3 4 5 6

Small cities’

residents30.200 0.207 0.188 0.207 0.233 0.209

Countryside’s

residents30.180 0.228 0.357 0.301 0.267 0.318

Married children 0.664 0.643 0.587 0.593 0.613 0.628

Children living

apart from parents

0.276 0.150 0.166 0.188 0.304 0.326

N of members in

the children’s

household

4.176

[1.769]

4.317

[1.614]

3.877

[1.404]

4.099

[1.574]

N of children aged

under 3 in the

children’s

household

0.181

[0.351]

0.199

[0.362]

0.204

[0.370]

0.251

[0.382]

N of children aged

under 17 in the

children’s

household

0.608

[0.775]

0.998

[0.779]

0.618

[0.708]

1.070

[0.745]

Children’s

education

below-secondary or

secondary

0.224 0.250 0.377 0.214 0.271 0.138

Children with

technical school

0.404 0.442 0.390 0.396 0.400 0.335

Tertiary children’s

education

0.372 0.308 0.232 0.390 0.329 0.527

Russian,

Ukrainian, or

Belorussian

children

0.924 0.899 0.800 0.880 0.888 0.908

30

Sample 1 2 3 4 5 6

Christian

(non-Russian)

children

0.028 0.022 0.040 0.020 0.033 0.025

Moslem children 0.032 0.059 0.130 0.075 0.037 0.050

Children of other

nationalities

0.016 0.020 0.030 0.025 0.042 0.017

The Central

region’s residents

0.308 0.322 0.252 0.270 0.288 0.331

The North Western


0.072 0.083 0.086 0.086 0.075 0.075

The Volga region’s

residents

0.292 0.247 0.205 0.220 0.233 0.239

The Southern


0.116 0.125 0.123 0.137 0.125 0.109

The North

Caucasian region’s

residents

0.028 0.048 0.112 0.059 0.029 0.029

The Ural region’s

residents

0.072 0.059 0.055 0.072 0.075 0.063

The Siberian


0.100 0.098 0.109 0.099 0.117 0.100

The Far Eastern


0.012 0.018 0.058 0.057 0.058 0.054

Children employed

as army

0.003 0.008 0.001

Children employed

as offi cials,

managers, and

professionals

0.244 0.190 0.287

Children employed

as technicians

0.205 0.106 0.286

31

Sample 1 2 3 4 5 6

Children employed

as clerks and

service workers

0.179 0.077 0.263

Children employed

as craft and related

trades

0.125 0.226 0.037

Children employed

as machine

operators and

assemblers

0.143 0.258 0.035

Children employed

as skilled

agricultural,

fishery and

unskilled workers

0.101 0.135 0.091

Number of

observations with

non-missing

children’s

occupation

733 726 696

1Sample means with standard deviations in brackets and shares of spe-cified groups.

2The sample mean of residuals of the target group from equation (8)

averaged over 5 years or more.3The settlements situated in the Moscow’s administrative region (oblast’)

are classified as capital cities while estimating equations (1)-(3). They are

classified according to their actual population while estimating equation (4)

and equation (5).

32

Table 6. The selected estimates of the wage equation by the Heckmanselection estimator.

Log of the real wage

index

0.8711*** [0.0243]

Log of hours monthly

worked at the

primary place

0.3519*** [0.0111]

Constant 6.2933*** [0.0716]

rho 0. 2887 [0.0140]

Wald test of

independent

equations (rho = 0)

chi2(1) = 378.95 ***

Number of

observations

77016

Censored observations 2540

Uncensored

observations

74476

The dependent variable of the wage equation is log of deflated monthly

monetary earnings received at primary place of work. The observation is

selected if earnings are reported, and not selected if the employed individual

explicitly refuses to declare earnings.


Robust standard errors adjusted for 20628 clusters in individuals are in

brackets.

33

Table 7. The OLS estimates of equation (1).

Parent’s earnings1 0.3310***

[0.0583]

The Southern

region

0.0367

[0.1174]

More-paid (1) or

solely available (0)

parent is selected

0.0442

[0.0680]

The North

Caucasian region

-0.0832

[0.2301]

Parent’s gender -0.0207

[0.0663]

The Ural region -0.0306

[0.1327]

Big cities2 0.1431

[0.1351]

The Siberian

region

-0.1779

[0.1175]

Medium-sized

cities

-0.1673

[0.1143]

The Far Eastern

region

-0.3464

[0.2238]

Small towns -0.2233*

[0.1302]

Constant 0.1958**

[0.0894]

The countryside -0.3325**

[0.1328]

Number of

observations

250

The North

Western region30.2392*

[0.1285]

R-squared 0.3062

The Volga region -0.2500**

[0.1037]

The dependent variable is the individual mean of the children’s residuals

from equation (8) averaged over no less than 5 rounds of the survey.1 The individual mean of the parent’s residuals from equation (8) averaged

over no less than 5 rounds of the survey.2 The reference group is capital cities including localities from the Mo-

scow’s region.3 The reference group is the Central region.

* Significant at 0.1 level.

** Significant at 0.05 level.

*** Significant at 0.01 level.

Robust standard errors adjusted for 233 clusters in parents are in brackets.

34


Parent’s

earnings10.2542***

[0.0610]

Medium-sized

cities

-0.1375

[0.0948]

Parent’s

technical school20.0532

[0.0621]


[0.1081]

Tertiary

parent’s

education

0.1236

[0.0904]

The countryside -0.3607***

[0.1167]

More-paid (1) or

solely available

(0) parent is

selected

-0.0242

[0.0690]

The North

Western region50.2871***

[0.1072]

Parent’s gender 0.0278

[0.0641]

The Volga

region

-0.2050**

[0.0829]

Children’s

technical school20.1337**

[0.0624]

The Southern

region

0.0332

[0.1110]

Tertiary

children’s

education

0.3004***

[0.0809]

The North

Caucasian

region

-0.0945

[0.2375]

Children’s

gender

0.3840***

[0.0563]

The Ural region 0.0440

[0.1203]

Children’s

marital status

0.1156**

[0.0543]

The Siberian

region

-0.1861

[0.1146]

Christian

(non-Russian)

children3

-0.0560

[0.2974]

The Far Eastern

region

-0.2254

[0.2703]

Moslem children -0.0221

[0.1148]

Constant -0.2856**

[0.1107]

Children of

other

nationalities

0.0460

[0.1875]

Number of

observations

250

35

Big cities4 0.0961

[0.1164]

R-squared 0.4695



over no less than 5 rounds of the survey.2 The reference group is secondary education or less.3 The reference group is Russian, Ukrainian, and Belorussian.4 The reference group is capital cities including localities from the Mo-

scow’s region.5 The reference group is the Central region.




Robust standard errors adjusted for 233 clusters in parents are in brackets.

36

Table 9. The selected estimates of equation (2) by the quantile regression.

Quantiles

0.20 0.40 0.60 0.80

Parent’s

earnings10.3184***

[0.1048]

0.2991***

[0.0792]

0.2300**

[0.0891]

0.3823***

[0.0949]

Parent’s

technical school2-0.0671

[0.1057]

-0.0182

[0.0907]

0.0348

[0.0836]

0.1055

[0.0951]

Tertiary

parent’s

education

0.0328

[0.1583]

0.0546

[0.1282]

0.0873

[0.1255]

-0.0089

[0.1205]

Children’s


[0.0924]

0.1617*

[0.0872]

0.1755**

[0.0893]

0.1066

[0.1068]

Tertiary

children’s

education

0.3676***

[0.1255]

0.3471***

[0.1049]

0.4027***

[0.1038]

0.3447***

[0.1152]

Number of

observations

250 250 250 250

Pseudo

R-squared

0.3252 0.3285 0.3058 0.3071



over no less than 5 rounds of the survey.2 The reference group is secondary education or less.




Bootstrap standard errors are in brackets.

37


Parent’s technical

school10.0067

[0.0395]

The North

Western region40.2799***

[0.0623]

Tertiary parent’s

education

0.0691

[0.0504]

The Volga region -0.0862

[0.0655]

More-paid (1) or

solely available (0)

parent is selected

0.0467

[0.0531]

The Southern

region

0.0814

[0.0675]

Parent’s gender 0.0238

[0.0353]

The North

Caucasian region

-0.0282

[0.0937]

Children’s


[0.0434]

The Ural region 0.1532*

[0.0928]

Tertiary children’s

education

0.2947***

[0.0548]

The Siberian

region

-0.0112

[0.0750]

Children’s gender 0.3445***

[0.0383]

The Far Eastern

region

0.2477

[0.1705]

Children’s marital

status

0.1091***

[0.0330]

Children employed

as army50.0560

[0.1446]

Christian

(non-Russian)

children2

0.0570

[0.1294]

Children employed

as technicians

-0.0719

[0.0508]

Moslem children 0.0699

[0.0772]

Children employed

as clerks and

service workers

-0.1734***

[0.0567]

Children of other

nationalities

0.1150

[0.1552]

Children employed

as craft and related

trades

-0.0672

[0.0720]

Big cities3 -0.1088

[0.0791]

Children employed

as machine

operators and

assemblers

-0.0856

[0.0690]

38

Medium-sized

cities

-0.2434***

[0.0540]

Children employed

as skilled

agricultural,

fishery and

unskilled workers

-0.4002***

[0.0657]

Small towns -0.3240***

[0.0741]

Constant -0.0393

[0.0715]


[0.0660]

Number of

observations

733

R-squared 0.4178


from equation (8) averaged over no less than 5 rounds of the survey.1 The reference group is secondary education or less.2 The reference group is Russian, Ukrainian, and Belorussian.3 The reference group is capital cities including localities from the Mo-

scow’s region.4 The reference group is the Central region.5 The reference group is offi cials, managers, and professionals.




Robust standard errors adjusted for 681 clusters in parent are in brackets.

39

Table 11. The estimates of the ordered probit model on the sample of sonsand fathers (equation (4)).

Father’s

technical school10.5317***

[0.0929]

The Siberian

region

-0.4054***

[0.1576]

Tertiary father’s

education

0.9509***

[0.1428]

The Far Eastern

region

-0.2954

[0.2141]

Son’s marital

status

0.1511

[0.1009]

Son born in

1976-19805-0.3608***

[0.1220]

Christian

(non-Russian)

son2

-0.2054

[0.1987]

Son born in

1971-1975

-0.4684***

[0.1340]

Moslem son -0.4107**

[0.2021]

Son born in

1966-1970

-0.3619**

[0.1578]

Son of other

nationalities

0.5745***

[0.2150]

Son born in

1961-1965

-0.2877

[0.1830]

Big cities3 0.3159

[0.2128]

Son born in

1956-1960

-0.3053

[0.2369]

Medium-sized

cities

0.2581

[0.1723]

Son born in

1947-1955

-0.5856

[0.3676]

Small towns 0.0572

[0.1789]

N of members in

the son’s

household

-0.0358

[0.0294]

The countryside -0.2301

[0.1719]

N of children

aged under 3 in

the son’s

household

-0.0472

[0.1499]

The North

Western region4-0.2610

[0.1750]

N of children

aged under 17

in the son’s

household

0.1139

[0.0808]

The Volga

region

-0.3926***

[0.1449]

Cutpoint 1 -0.5206

[0.2234]

40

The Southern

region

0.0187

[0.1545]

Cutpoint 2 0.6694

[0.2228]

The North

Caucasian

region

0.1750

[0.2209]

Number of

observations

848

The Ural region -0.5818***

[0.2015]

Pseudo

R-squared

0.0925

The dependent variable is the 3-level categorical variable of son’s educa-

tion attainment.1 The reference group is secondary education or less.2 The reference group is Russian, Ukrainian, or Belorussian.3 The reference group is capital cities excluding localities from the Mo-

scow’s region.4 The reference group is the Central region.5 The reference group is sons born in 1981-83 years.

** Significant at .05 level.


Robust standard errors adjusted for 742 clusters formed by father are in

brackets.

41

Table 12. The estimates of the ordered probit model on the sample ofdaughters and fathers (equation (4)).

Father’s


[0.0962]

The Siberian

region

-0.4364**

[0.1795]

Tertiary father’s

education

0.9672***

[0.1390]

The Far Eastern

region

-0.3046

[0.2099]

Daughter’s

marital status

0.1092

[0.0972]

Daughter born

in 1976-19805-0.2439**

[0.1196]

Christian

(non-Russian)

daughter2

0.4711*

[0.2687]

Daughter born

in 1971-1975

-0.3474***

[0.1288]

Moslem

daughter

-0.3246*

[0.1748]

Daughter born

in 1966-1970

-0.2689*

[0.1493]

Daughter of

other

nationalities

-0.1919

[0.2725]

Daughter born

in 1961-1965

-0.1971

[0.1749]

Big cities3 0.2504

[0.2155]

Daughter born

in 1956-1960

0.0527

[0.2891]

Medium-sized

cities

-0.2014

[0.1717]

Daughter born

in 1947-1955

-0.5885**

[0.2826]


[0.1974]

N of members in

the daughter’s

household

-0.0908***

[0.0306]


[0.1744]

N of children

aged under 3 in

the daughter’s

household

0.2438*

[0.1311]

The North

Western region4-0.2224

[0.1718]

N of children

aged under 17 in

the daughter’s

household

-0.0769

[0.0718]

42

The Volga

region

-0.2137

[0.1462]

Cutpoint 1 -1.5702

[0.2180]

The Southern

region

0.1521

[0.1580]

Cutpoint 2 -0.3243

[0.2123]

The North

Caucasian

region

-0.2400

[0.2355]

Number of

observations

803

The Ural region -0.5359***

[0.1822]

Pseudo

R-squared

0.1124

The dependent variable is the 3-level categorical variable of daughter’s

education attainment.1 The reference group is secondary education or less.2 The reference group is Russian, Ukrainian, or Belorussian.3 The reference group is capital cities excluding localities from the Mo-

scow’s region.4 The reference group is the Central region.5 The reference group is daughters born in 1981-83 years.

* Significant at .10 level.

** Significant at .05 level.


Robust standard errors adjusted for 734 clusters formed by father are in

brackets.

43

Table 13. The selected estimates of the ordered probit model on the dataon son-father and daughter-father pairs (equation (5)).

Son-father

pairs

Daughter-

father

pairs

Father’s

earnings1-0.0335

[0.1588]

0.0968

[0.1872]

Father’s


[0.1678]

0.7605***

[0.2046]

Tertiary father’s

education

0.7932***

[0.2719]

1.4000***

[0.2865]

Cutpoint 1 -1.0879

[0.3941]

-2.3689

[0.5290]

Cutpoint 2 0.1984

[0.3949]

-1.0611

[0.5017]

Number of

observations

240 239

Pseudo

R-squared

0.1438 0.1817

The dependent variable is the 3-level categorical variable of children’s

education attainment.1 The individual mean of the predicted father’s residuals from equation

(8) averaged over no less than 5 rounds of the survey.2 The reference group is secondary education or less.


Robust standard errors are in brackets. The standard errors estimated

on the son-father data are adjusted for 206 clusters formed by father. The

standard errors estimated on the daughter-father data are adjusted for 207

clusters formed by father.

44

the intergenerational transmission of human capital and ...ftp.iza.org/dp10300.pdf · discussion...

Documents